Yuanta Securities’ CIO Office said NVIDIA’s expanded partnership with MediaTek marks a broader push from edge AI into data-center infrastructure, device-based AI computing and automotive platforms. The brokerage said NVIDIA’s US$3.5 billion investment in MediaTek convertible bonds underlines a longer-term commitment to the collaboration.
Yuanta said the most significant new element is in data centers, where MediaTek will join NVIDIA’s NVLink Fusion ecosystem. Under the arrangement, MediaTek will use NVIDIA technology to support the design of custom XPUs for hyperscalers, cloud service providers and frontier AI model developers.
The move could allow customers to develop their own custom AI chips while linking those processors into NVIDIA’s AI infrastructure, Yuanta said.
At the center of the structure is NVLink Fusion, which enables custom XPUs to connect with NVIDIA’s NVLink scale-up fabric without requiring customers to build a separate interconnect system from scratch. NVLink-C2C is used to connect custom XPUs with NVIDIA CPUs inside the same system, allowing customer-designed silicon to operate with NVIDIA’s architecture from the chip level through to rack-scale systems.
Yuanta said the partnership also extends into memory and advanced packaging through NVHBM. The technology incorporates NVIDIA-designed HBM base die and memory controller capabilities, enabling NVIDIA and MediaTek to support customers across compute design, connectivity, memory architecture and chip integration into rack-scale systems using NVIDIA-defined interfaces.
Beyond data centers, the two companies will continue existing work in AI devices and automotive platforms. MediaTek previously contributed to the SoC for the GB10 Grace Blackwell Superchip used in DGX Spark. Yuanta said MediaTek will also continue co-developing chips for RTX Spark and DGX Spark over multiple future generations.
In automotive, NVIDIA and MediaTek will keep working on platforms for AI-powered software-defined vehicles, according to Yuanta.
On investment implications, Yuanta said it has a positive view on NVIDIA. The brokerage said NVLink Fusion gives customer-developed custom XPUs a route into NVIDIA’s infrastructure, meaning NVIDIA can still sell supporting technologies even if hyperscalers and AI labs increasingly build their own accelerators.
Yuanta said those technologies include NVLink, NVLink switches, CPUs, networking products and rack architecture. This could help NVIDIA maintain a key position in AI infrastructure even if custom ASICs replace GPUs in some workloads.
Yuanta also identified potential beneficiaries in custom ASICs and connectivity. MediaTek, Marvell Technology and Astera Labs are directly involved in the NVLink Fusion ecosystem, it said. MediaTek and Marvell could benefit from helping customers design custom XPUs that connect to NVLink, while Astera Labs could gain from connectivity demand inside scale-up racks linking multiple XPUs.
For Broadcom, Yuanta described the impact as mixed. The company may benefit from the expansion of custom ASICs and continued growth in AI networking, but NVLink Fusion also gives customers another way to combine custom silicon with scale-up architecture through NVIDIA’s ecosystem. Yuanta said that competes with the Ethernet scale-up approach Broadcom is promoting.
Yuanta said a broader range of custom XPU designs would increase complexity in memory and packaging. That would be positive for Synopsys and Cadence Design Systems because of higher demand for EDA and verification work, it said.
TSMC could benefit from advanced process nodes and CoWoS packaging, Yuanta added. Samsung may also have additional upside from NVHBM because it is a technology partner in the NVLink Fusion ecosystem and has HBM, logic/foundry and packaging capabilities within the same company.
However, Yuanta noted that NVIDIA has only said NVHBM will work with leading memory makers and has not confirmed Samsung as a direct NVHBM manufacturer.
Yuanta said the collaboration will continue across future RTX Spark and DGX Spark generations, while NVIDIA and MediaTek will also keep developing automotive platforms for AI-powered software-defined vehicles.
